AI · AI today, November 5, 2026
Models
Baidu releases ERNIE 5.5 Flash Base, a 280-billion-parameter open model
· Aster editorial team
AI-generated imageBaidu released ERNIE 5.5 Flash Base, a language model whose files anyone can download from Hugging Face. It's a “base” model: it ships without the extra training that turns it into an assistant, so each team can adapt it to its own needs. Baidu says it competes with similarly sized open models such as DeepSeek V4 Flash and MiMo V2.5 on reasoning, coding and long texts. A fine-tuned version will follow later.
Why it matters
Every open model coming out of China gives companies and developers another way to run AI on their own hardware without paying a provider per use. Releasing the raw base version gives more freedom to anyone who wants to specialize it.
For the technical crowd
Mixture-of-experts (MoE) architecture: 280 billion total parameters, 14 billion active per query. Trained on 33.5 trillion tokens with a training context of up to 128,000 tokens; designed to handle up to 1 million tokens during fine-tuning and inference. Weights in FP8 and BF16.
· Source: The Decoder
More from AI today
- OpenAI launches GPT-6 Cyber, a model that has already found 81 serious flaws in FirefoxOpenAI introduced GPT-6 Cyber, a model built to find and exploit security flaws in software. For now, only approved customers in its Daybreak Red…
- Greg Brockman says OpenAI could be ready to go public in 2027OpenAI president Greg Brockman answered “I think so. Yes” when asked whether the company would be ready for an IPO next year. He spoke at a Wall…
- Meta warns post-election backlash against data centers will slow AI expansionMeta president Dina Powell McCormick said the results of the US midterm elections will make it harder to build data centers. She acknowledged the…
- Google launches Gemini Omni 1.2 Flash: AI videos up to 15 seconds long and 30% cheaperGoogle introduced Gemini Omni 1.2 Flash, a model that creates and edits videos from text, images or audio. Clips can now run up to 15 seconds, up…
- Tencent launches Hy4, its first model that sees images and operates software on its ownTencent unveiled Hy4, a language model that understands images and can operate software by itself, something its earlier versions couldn't do. It's…